Grading and submission

Seeing where you stand

Each stage scores itself as it runs and writes the result to disk, so you can ask for your standing at any time without rebuilding anything:

python prng_build.py --grades

Every check names itself, says what it is worth, and — when it fails — says why:

Python LFSR model: 10/10
  ✓ (4/4) Generated at least 1000 samples
  ✓ (3/3) Values are in range 0 to 255
  ✓ (3/3) The generator does not get stuck
Evaluating the generator: 7/10
  ✓ (2/2) Your lag-1 correlation is correct
  ✗ (0/3) Successive samples are uncorrelated — lag-1 correlation = +0.5042,
          which is 15.9 standard errors from zero — each sample predicts the
          next far too well
  ✓ (3/3) Your histogram is correct
  ✓ (2/2) The histogram is flat

This output is the authority on the marking scheme, which is why these pages do not repeat it — a page can go stale, the build cannot.

A stage you have not run yet shows as not yet run and scores zero.

The build only redoes what changed

The build tracks file timestamps. Edit prng_model.py and the Python stage re-runs — and so do evaluation and simulation, because both work from its output. Edit only prng_next.sv and just the simulation re-runs. Change nothing and every stage reports UP-TO-DATE and does no work.

If you ever want to force everything to run again:

python prng_build.py --force

Building the submission

submit is a build step like any other, and it is the last one in the graph:

python prng_build.py --through submit
python prng_build.py                     # the same thing — submit is the default

Because it is a step, it depends on the three graded stages, so running it runs whatever is out of date first. There is no separate script to remember and no way to bundle results without regenerating them.

It writes:

submission/submitted_results.json    your scores and feedback
submission/submission.zip            what you upload

The zip contains that results file, your four source files, vectors/prng_py.csv, and the histogram you drew in stage 2. Upload submission.zip to Gradescope.

Your figure is in there so it can be looked at. Nothing checks what it shows — only that you produced one — but it is the part of this lab whose quality a person can judge at a glance and a script cannot judge at all.

Gradescope reports the score in the zip you upload. Run the full build after your last edit — if you fix something and upload without re-running, you submit the old score.

You can submit an unfinished lab. A stage you never ran scores zero and the zip is still built, so partial work is always submittable.

What is graded, and what is not

Graded	The numbers your code produces, whether your measurements of them are right, and whether your two implementations agree
Not graded	Which algorithm you used, your variable names, your style

Nothing is compared against a stored reference sequence. Any correct PRNG passes the Python stage, and the SystemVerilog stage checks it against your Python rather than against ours. The write-up pins one algorithm so that nobody has to guess what to build — not because the grader insists on it.

The one thing this does not let you do is leave both sides unimplemented. Two untouched placeholders agree with each other perfectly, so a constant stream scores zero on the comparison no matter how well it matches. See the SystemVerilog stage.